Europe leads the world on AI rules. So why is it missing from the AI safety debate?

Nicolas Economou/NurPhoto/Getty Images

Europe has been conspicuously absent from high-level talks about how to use artificial intelligence safely, despite crafting the world’s first comprehensive set of rules to govern the technology.

The irony underscores how far Europe has fallen behind the United States and China in the AI race. Although Europe took the lead on AI regulation years ago, it has fewer frontier models, data centers and big tech companies to compete on an equal footing with its larger trading partners.

“Europe is at acute risk of marginalization on the global stage because of the extraordinary developments of AI,” warned Nobel-prize winning French economist Philippe Aghion. He co-wrote recent report that argues Europe needs an urgent “course correction if it wants to thrive in the AI era.”

The European Union has long aspired to be the global leader in “safe AI,” with a strategy that includes hefty fines. Yet, as mostly American AI executives recently committed to “ self-police ” the world’s leading AI models, it was patently clear that the EU’s approach was not the one being followed.

Kevin Dietsch/Getty Images

Jonathan Ernst/Reuters

Europe has long struggled to produce the tech giants boasted by Silicon Valley. The problem now extends to a lack of frontier AI models that can rival the likes of OpenAI’s ChatGPT, Anthropic’s Claude or comparable tools from China’s DeepSeek.

European Central Bank president Christine Lagarde last month described the continent’s AI predicament as “awkward.”

“Either (Europe) holds back on adopting, because it cannot protect its data, and forgoes the growth — or it adopts AI quickly, becomes highly dependent, and risks losing the freedom to organize its economy according to its own values,” she said in a speech .

Paris-based Mistral AI unveiled a model last week that it argues is on par with its better-known American competitors, providing a glimmer of hope for Europe’s AI industry, but it remains to be seen if it will provide a serious rival to offerings from the AI giants.

Ian Langsdon/AFP/Getty Images

Years before debates about how to police AI models became the topic du jour , European lawmakers drafted the AI Act to hold companies accountable for the risks arising from their tools, including strict requirements in areas for data quality, human oversight and cybersecurity.

Non-compliance can result in forced changes to tools or fines of up to 3% of annual turnover, or total income, which would amount to billion of dollars for the biggest tech companies.

The AI Act is “a crucial piece to putting guardrails in place. And it puts Europe in the position to shape global efforts on how to tackle” the dangers of the technology, European Commission President Ursula von der Leyen said in her state of the union address last month.

But in practice, the law has proven difficult to enforce, with several EU member states yet to establish authorities to police compliance.

The law, alongside other tech regulations like the Digital Markets Act and the General Data Protection Regulation, has also contributed to delays in the rollout of AI products in Europe.

Martin Bertrand/Hans Lucas/AFP/Getty Images

Meta AI, for example, arrived in Europe in March 2025, some 18 months after its US rollout began.

“It’s taken longer than we would’ve liked get our AI technology into the hands of people in Europe as we continue to navigate its complex regulatory system,” Meta said at the time.

Regulation holds back European companies, too. Roughly half of 157, 000 firms surveyed last year by Eurostat, the EU statistics office, cited a lack of clarity about legal consequences and concerns about violating the bloc’s stringent data protection laws as reasons not to adopt AI.

The AI Act and related national regulations are “slowing us down,” said Eléonore Crespo, the co-founder of Pigment, a Paris-based firm valued at more than $1 billion that builds AI software for businesses.

Europe’s economy needs “innovative companies” to survive. “If regulation holds us back, it’s not going to work,” she told CNN.

Chris J. Ratcliffe/Bloomberg/Getty Images

The European Commission has itself tacitly acknowledged the regulatory burden. The AI Omnibus , a package of legislation introduced last year, watered down some requirements of the AI Act and extended the timeline for the implementation rules relating to high-risk AI systems to December 2027.

Legal experts and AI researchers say it is too soon to judge whether the AI Act has been a success, partly because the rules are still being phased in. But doubts remain about whether the legislation can keep pace with AI’s rapid-fire development.

“The prohibitions (of the act) are broadly sound and the transparency requirements are relatively straightforward,” said Amir Banifatemi, the chief responsible AI officer at Cognizant AI lab, a research firm based in San Francisco. “The high-risk regime is the weak point, and it has not yet begun to apply,” he told CNN.

Banifatemi pointed to AI agents as one potential problem. He said agents may combine tools from different providers, raising thorny problems over who’s accountable if something goes wrong.

Regulation is far from the only barrier to greater AI uptake in Europe. A lack of investment in infrastructure and private companies may be an even bigger problem.

Private AI investment in Europe totaled roughly $21 billion last year, a fraction of the $285.9 billion invested in the United States, according to the Stanford Institute for Human-Centered Artificial Intelligenc e.

The center’s recent AI Index Report shows the United States hosts 10 times more data centers than any other country and produced 59 notable AI models in 2025, compared to China’s 35 and just one in each of France and the United Kingdom.

Yang: Top A.I. executives self-policing ‘worst outcome’ after calls slowdown and regulation

Europe is in the “middle ages” when it comes to AI, Crespo quipped. But she remains optimistic that it will eventually have its “renaissance.”

“I’m quite positive about the potential because I see some very cool projects,” she said, pointing to the emergence of more AI companies and tools on the continent as well as a depth of technical talent.

Other experts are also optimistic, arguing that Europe’s industrial might and high-quality data is a distinct advantage when it comes to exploiting real-world applications for AI.

“Europe has deep industrial expertise and valuable domain data in manufacturing, energy, mobility and healthcare,” Banifatemi said.

“That is where much of AI’s economic value may ultimately be created.”

Nicolas Economou/NurPhoto/Getty Images

尽管欧洲制定了世界上第一套全面的人工智能技术管理规则，但在有关如何安全使用人工智能的高级别会谈中，欧洲却明显缺席。

讽刺的是，这凸显了欧洲在人工智能竞赛中与美国和中国相比落后多少。尽管欧洲多年前就已在人工智能监管方面处于领先地位，但它拥有的前沿模型、数据中心和大型科技公司数量较少，难以与规模更大的贸易伙伴展开公平竞争。

“由于人工智能的飞速发展，欧洲在全球舞台上面临着被边缘化的严重风险，”诺贝尔经济学奖得主、法国经济学家菲利普·阿吉翁警告说。他与人合著的一份最新报告指出，如果欧洲想要在人工智能时代蓬勃发展，就需要紧急进行“方向调整”。

欧盟长期以来一直致力于成为“安全人工智能”领域的全球领导者，其战略包括巨额罚款。然而，鉴于近期主要由美国人工智能企业高管组成的联盟承诺对全球领先的人工智能模型进行“自我监管”，显然，欧盟的做法并未得到广泛应用。

Kevin Dietsch/Getty Images

乔纳森·恩斯特/路透社

欧洲长期以来一直难以培育出像硅谷那样的科技巨头。如今，问题还延伸到缺乏能够与OpenAI的ChatGPT、Anthropic的Claude或中国DeepSeek的同类工具相媲美的前沿人工智能模型。

欧洲中央银行行长克里斯蒂娜·拉加德上个月将欧洲大陆的人工智能困境描述为“尴尬的”。

她在一次演讲中说：“要么（欧洲）因为无法保护自己的数据而迟迟不愿采用人工智能，从而放弃增长；要么它迅速采用人工智能，变得高度依赖人工智能，并有可能失去根据自身价值观组织经济的自由。”

总部位于巴黎的 Mistral AI 上周发布了一款模型，该公司声称该模型与更为知名的美国竞争对手不相上下，这为欧洲人工智能产业带来了一线希望，但它能否对人工智能巨头的产品构成真正的竞争，还有待观察。

伊恩·朗斯顿/法新社/盖蒂图片社

早在关于如何监管​​人工智能模型的辩论成为热门话题之前，欧洲立法者就起草了《人工智能法案》，以追究公司对其工具带来的风险的责任，其中包括对数据质量、人工监督和网络安全等领域的严格要求。

不遵守规定可能导致强制更改工具或处以相当于年营业额或总收入 3% 的罚款，对于最大的科技公司来说，罚款金额将高达数十亿美元。

欧盟委员会主席乌尔苏拉·冯德莱恩在上个月的国情咨文中表示，《人工智能法案》是“建立安全保障措施的关键一环。它使欧洲能够在应对这项技术带来的危险方面发挥全球引领作用”。

但实际上，这项法律很难执行，一些欧盟成员国尚未设立机构来监督执行情况。

该法律以及其他科技法规，如《数字市场法》和《通用数据保护条例》，也导致了人工智能产品在欧洲推广的延迟。

Martin Bertrand/Hans Lucas/法新社/Getty Images

例如，Meta AI 于 2025 年 3 月进入欧洲市场，比其在美国推出晚了大约 18 个月。

Meta当时表示：“由于我们仍在努力适应欧洲复杂的监管体系，因此将我们的人工智能技术推广到欧洲人民手中所花费的时间比我们预期的要长。”

监管也阻碍了欧洲企业的发展。欧盟统计局（Eurostat）去年对15.7万家企业进行的调查显示，约有一半的企业表示，由于缺乏关于法律后果的明确信息以及担心违反欧盟严格的数据保护法，他们不愿采用人工智能。

“人工智能法案及相关国家法规正在拖慢我们的步伐，”总部位于巴黎、估值超过 10 亿美元的人工智能软件开发公司 Pigment 的联合创始人 Eléonore Crespo 表示。该公司为企业开发人工智能软件。

欧洲经济需要“创新型企业”才能生存。“如果监管阻碍了我们，那就行不通，”她告诉CNN。

克里斯·J·拉特克利夫/彭博社/盖蒂图片社

欧盟委员会自身也已默认了监管负担。去年推出的《人工智能综合法案》是一揽子立法，它淡化了《人工智能法》的部分要求，并将高风险人工智能系统相关实施规则的期限延长至2027年12月。

法律专家和人工智能研究人员表示，现在判断《人工智能法案》是否成功还为时过早，部分原因是相关规则仍在分阶段实施。但人们仍然怀疑该法案能否跟上人工智能的飞速发展。

“该法案的禁令总体上是合理的，透明度要求也相对简单明了，”旧金山研究公司Cognizant AI实验室的首席人工智能责任官阿米尔·巴尼法特米表示。“高风险机制是薄弱环节，而且尚未开始实施，”他告诉CNN。

巴尼法特米指出人工智能代理可能存在一个潜在问题。他说，代理可能会整合来自不同供应商的工具，一旦出现问题，责任归属问题就会变得棘手。

监管远非欧洲人工智能普及应用的唯一障碍。基础设施和私营企业投资不足或许才是更大的问题。

根据斯坦福大学以人为本人工智能研究所的数据，去年欧洲私人人工智能投资总额约为 210 亿美元，仅为美国 2859 亿美元投资额的一小部分。

该中心最近发布的AI指数报告显示，美国的数据中心数量是其他任何国家的10倍，并且到2025年将产生59个值得关注的AI模型，相比之下，中国有35个，法国和英国各只有一个。

杨：人工智能高管呼吁放缓发展速度并加强监管，之后他们开始自我监管，以避免“最坏结果”。

克雷斯波打趣道，欧洲在人工智能领域还处于“中世纪”。但她仍然乐观地认为，人工智能最终会迎来“复兴”。

“我对人工智能的潜力非常乐观，因为我看到了一些非常酷的项目，”她说道，并指出非洲大陆涌现出更多人工智能公司和工具，以及深厚的技术人才储备。

其他专家也持乐观态度，认为欧洲的工业实力和高质量数据在利用人工智能的现实应用方面具有明显的优势。

“欧洲在制造业、能源、交通运输和医疗保健领域拥有深厚的工业专业知识和宝贵的领域数据，”巴尼法特米说。

“人工智能的大部分经济价值最终可能就产生于此。”